mpep.io

The USPTO Manual of Patent Examining Procedure as structured, citable data — for LLM agents.

Query every MPEP section, statute, rule, and case over a clean REST API or a hosted MCP server. Answers are grounded in a pinned, hash-verified corpus, so your agent cites what the manual actually says instead of hallucinating it.

Get a free API key → · 60 req/min, 5,000/day · REST + MCP + llms.txt

What's in it

• All 736 MPEP sections (Ninth Edition, R-01.2024), full text with cited authorities.
• 1,728 cases with AI-generated holding summaries and full opinion text.
• 35 U.S.C. and 37 CFR provision text, and examiner form paragraphs.
• The citation graph — resolve any citation, and reverse-lookup every section that cites a case or statute.

Everything is pinned to one hash-verified corpus revision, so the same query returns the same grounded answer every time.

Quickstart

Python (mpep · zero dependencies):

pip install mpep
export MPEP_API_KEY=...        # from https://mpep.io/keys

from mpep import Client
mpep = Client()
hits = mpep.search("motivation to combine")
print(mpep.resolve_citation("KSR")["holding"])

TypeScript (@mpep/client · zero dependencies):

npm install @mpep/client

import { MPEP } from "@mpep/client";
const mpep = new MPEP({ apiKey: process.env.MPEP_API_KEY });
const { results } = await mpep.search("motivation to combine");

MCP (Claude, Cursor, VS Code, …) — one line, no install:

claude mcp add mpep --transport http https://api.mpep.io/mcp \
  --header "Authorization: Bearer YOUR_KEY"

See mcp/ for every client's config.

curl:

curl -H "Authorization: Bearer YOUR_KEY" \
  "https://api.mpep.io/v1/search?q=obviousness"

Tools / endpoints

Tool (MCP)	REST	Does
search	GET /v1/search	full-text search across every section
get_section	GET /v1/sections/{n}	section text + cited authorities
resolve_citation	GET /v1/citations/resolve	resolve a cite; statute text / case holding
what_cites	GET /v1/what-cites	reverse citations into the manual
read_case	GET /v1/cases/{c}/opinion	full opinion text
get_form_paragraph	GET /v1/form-paragraphs/{n}	examiner form paragraph
coverage	GET /v1/coverage	edition/revision + coverage

• OpenAPI spec: openapi.json · live docs at api.mpep.io/docs
• Function-calling schemas (OpenAI / Anthropic / frameworks): tools.json
• Runnable example: examples/office_action.py

This repo vs. the API

This repo is the open-source client kit — SDKs, MCP configs, tool schemas, and examples — and it's MIT licensed; use and fork it freely. It talks to the hosted mpep.io API, where the structured data lives.

The underlying government texts (MPEP, USC, CFR, court opinions) are public domain. The structured representation, annotations, and AI holdings served by the API are proprietary to Patent Agentic LLC — see the terms. Not an official USPTO resource.

Links

mpep.io · Get a key · For agents · API docs